K-ballo changed the topic of #ste||ar to: STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar.cct.lsu.edu | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | Buildbot: http://rostam.cct.lsu.edu/ | Log: http://irclog.cct.lsu.edu/
<jaafar> I was browsing the code and noticed that there is a transform_loop_n that is not implemented in terms of loop_n - and the latter seems to have optimizations for e.g. SIMD
<jaafar> Should transform_loop_n use loop_n?
<hkaiser> jaafar: good question
K-ballo has quit [Quit: K-ballo]
hkaiser has quit [Quit: bye]
surbhi has joined #ste||ar
<gonidelis[m]> jaafar could you give a ref?
surbhi has quit [Ping timeout: 246 seconds]
surbhi has joined #ste||ar
K-ballo has joined #ste||ar
parsa has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
parsa has joined #ste||ar
parsa has quit [Client Quit]
parsa has joined #ste||ar
parsa has quit [Client Quit]
parsa has joined #ste||ar
parsa has quit [Client Quit]
parsa has joined #ste||ar
<jaafar> didn't use transform_loop_n when it was, in fact, a transform. So I replaced the code and it got slower
<jaafar> loop unrolling, which isn't in transform_loop_n
<jaafar> it seemed like rewriting transform_loop_n to use loop_n ought in theory to not cost anything, and possibly give gains for users
<gonidelis[m]> hmmm
<gonidelis[m]> jaafar: By taking this into consideration, we should optimize `transform_loop_n` accordingly.
<jaafar> It seemed to me that just using loop_n might be enough
<gonidelis[m]> jaafar: I happen to work in the `transform` C++20 adaptation right now and I have encountered `transform_loop_n` but it was just the interface
<gonidelis[m]> jaafar: should we not check into the performance of `transform_loop_n` though? Just to see where it stinks?
surbhi has quit [Ping timeout: 260 seconds]
<jaafar> oh sure, whatever you think :)
<jaafar> I was just surprised to see that one was not implemented in terms of the other. transform_loop_n has a raw C++ loop instead.
<gonidelis[m]> jaafar cool thanks for letting know... I might come across this issue in next few weeks
<gonidelis[m]> I will update in that case
diehlpk_work has joined #ste||ar